kth.sePublications
Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Physics-Informed Deep Learning for Transformer Thermal Modelling: Possibilities and Limitations
KTH, School of Engineering Sciences (SCI), Mathematics (Dept.), Mathematics (Div.).
2022 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Fysik-informerad djupinlärning för termisk modellering av transformatorer: möjligheter och begränsningar (Swedish)
Abstract [en]

Physics-Informed neural networks (PINNs) are a novel approach to integrate physical models with neural networks when solving supervised learning tasks. They have shown promising potential in solving partial differential equations (PDEs) using space and time coordinates as input. However despite their promise, they often fail to train when target PDEs contain high frequencies are multi-scale features.

Thermal modelling of power transformers is a fundamental practice for monitoring and improving their efficiency and lifetime.

In this work, we investigate the performance of different PINN models in a specific 1D thermal modelling problem derived to model the heat distribution inside a transformer. Measurements taken from a real transformer in service are used, which include the top-oil temperature, the ambient temperetaure, and the load factor. Using this example, we demonstrate the strengts and the limitations of PINNs, propose potential remedies, and provide an overall assessment of the future potential of applying PINNs to transformer thermal modelling. All code is publicly available at https://github.com/OliverOde/PINN_for_Transformer_Thermal_Modelling, accopanied with simulated fake data. 

Abstract [sv]

Fysik-informerade neurala nätverk (PNNs) är ett nytt tillvägagångssätt för att integrera fysikaliska modeller i neurala nätverk för att lösa problem inom väglett lärande (supervised learning). De har visat potential att kunna lösa partiella differentialekvationer genom att använda rum- och tidskoordinater som indata. Trots en lovande potential misslyckats de däremot ofta att lösa vissa typer av ekvationer, till exempel när lösningarna till dessa har höga frekvenser.

Termisk modellering av transformatorer är viktigt för att kunna övervaka och förbättra deras effektivitet and livstid.

I denna avhandling undersöker vi olika PINN-modellers prestation i ett specifikt 1D termiskt modelleringsproblem med syftet att modellera värmedistributionen inuti en transformator. Vi använder data som är tagna från en riktig transformator i arbete, som består av toppoljetemperatur, omgivande temperatur samt belastningsfaktor. Genom detta exempel demonstrerar visar vi PINNs styrkor och begränsningar, ger förslag på hur vi kan göra dem bättre, samt ger en bedömning av potentialen att använda PINNs för termisk modellering av transformatorer. All kod är tillgänglig här: https://github.com/OliverOde/PINN_for_Transformer_Thermal_Modelling,där datan som används har simulerats.

Place, publisher, year, edition, pages
2022. , p. 95
Series
TRITA-SCI-GRU ; 2022:378
Keywords [en]
ANN, PINN, PDEs, Transformers, Thermal Modelling
Keywords [sv]
ANN, PINN, PDEs, Transformatorer, Termisk modellering
National Category
Probability Theory and Statistics
Identifiers
URN: urn:nbn:se:kth:diva-354852OAI: oai:DiVA.org:kth-354852DiVA, id: diva2:1905716
External cooperation
Hitachi Energy
Subject / course
Mathematical Statistics
Educational program
Master of Science - Applied and Computational Mathematics
Supervisors
Examiners
Available from: 2024-10-15 Created: 2024-10-15 Last updated: 2024-10-15Bibliographically approved

Open Access in DiVA

fulltext(10227 kB)445 downloads
File information
File name FULLTEXT01.pdfFile size 10227 kBChecksum SHA-512
530d4e44f08640e85e40678f8be27afc793f97d0961fb8f1394ae7f9f3702403bc5585b1bf8cbd1eee150a11e9ccb3199acb373d8c242af1d60bd00eda36a1bd
Type fulltextMimetype application/pdf

By organisation
Mathematics (Div.)
Probability Theory and Statistics

Search outside of DiVA

GoogleGoogle Scholar
Total: 445 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 185 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf